Change within complex systems 

Most of the change methodologies we work with treat change as a tidy, controllable process: define the future state, plan a path to it, communicate, train, embed. Kotter, Prosci and plenty of others work this way¹ – perhaps revisiting stages en route, but mostly treating ‘change management’ as a set of separate steps that line up neatly with the project plan. And when ‘resistance’ turns up or adoption stalls, we lay the blame elsewhere – “There should have been more training” or “The comms could have been better”. 

While it’s certainly important to understand individuals’ emotional and psychological reactions to change and engage with them in ways that genuinely meets their needs, we know groups behave differently from people on their own, in many ways, and we must consider that.. Systems Thinking is one way of making sense of what’s going on². 

Feedback loops

Systems Thinking describes two kinds of feedback loop. 

• Negative or ‘balancing’ feedback loops: These correct themselves – they pull the system back towards the status quo. A thermostat is an example of a balancing feedback loop; when the temperature changes from the desired setting, heating or cooling kicks in to bring it back. 
• Positive or ‘reinforcing’ feedback loops: These feed themselves – they continue to push the system further from its starting point until it either collapses or something else steps in to stabilise it³. An epidemic is an example of a reinforcing feedback loop. 

Within organisations, feedback loops are extremely powerful, and a lot of what drives them is one very human habit: ‘conformity bias’, or our tendency to copy the people around us. There’s a lovely demonstration of this in the MusicLab experiments run by Salganik, Dodds and Watts⁴. They built an artificial music market and recruited over 14,000 people to listen to and download songs they’d never heard of. Everyone was randomly dropped either into an ‘independent’ group, with no visibility of other participants’ choices;  or into one of eight parallel ‘social influence’ worlds, where they could see what others had downloaded. The same 48 songs were available in every world; the same sorts of people took part in each; the only difference was the random popularity data at the start.

The results pointed to a significant level of social influence. When participants could see what their peers had chosen, the results were both more unequal and much harder to predict. Songs that did well in one world often did badly in another. The same song could finish first in one world and near the bottom in another. The best songs rarely did very badly, and the worst rarely did very well, but everything in between came down to luck. Small, random, early leads got amplified by social influence into wildly different end states – a textbook reinforcing loop, and one that happened under controlled conditions.

The same tendency explains why balancing feedback loops are so powerful, and why they’re the main reason human systems can be change-avoidant. Tools, reward structures and recruitment are just some of the things that reinforce ‘how things are done here’ and what’s expected of people. New joiners copy those around them, which is a big part of why culture usually beats individual personality. This was notoriously illustrated by the Milgram Experiments, in which 65% of participants were willing to inflict painful electric shocks simply because someone in authority told them to⁵. 

The good news for change practitioners is that people aren’t always bound by balancing loops. There’s an enduring body of work on ‘informational cascades’ suggesting that some kinds of conformity are surprisingly fragile⁶. When people are just going with the flow, without much information or much thought, their attitudes can shift quickly once conditions do. Everything looks settled until a new piece of public information turns up, and then the whole thing can flip from one behaviour to another. It’s the same pattern behind tipping points: long stretches where the balancing loops appear to have everything under control, broken up by rapid shifts when a reinforcing loop briefly takes charge and resets the system around something new⁷. 

Seen this way, successful organisational change means building a reinforcing loop strong enough to overcome the combined pull of all those balancing loops.

Balancing feedback loops – how systems self-correct 

‘Policy resistance’ is the term for the way systems respond to an intervention by adapting in ways that subtly defeat it⁸. The classic case is highway expansion: build more lanes to ease congestion, and within a few years the extra traffic they attract has filled them right back up⁹. The organisational version is the change programme that spawns a fresh set of workarounds, the new governance board that just adds delay, or the well-meant policy that produces exactly the behaviour it was designed to stop. 

Reinforcing feedback loops – how new paradigms replace old ones

Centola and Macy’s work focuses on what they call ‘complex contagion’¹⁰. Information spreads through a single contact (I tell you something, and now you know it). Behaviour works differently: most of us won’t change what we do until we’ve heard the message, or seen the example, from several people we trust who aren’t connected to each other. For a champions network, that means what looks like efficient ‘coverage’ (one champion per team, spread thinly) will build awareness but won’t shift behaviour. Changing behaviour needs overlap in the network – people need to see several colleagues they trust doing the new thing, more than once, before the new reinforcing loop is strong enough locally to shift the old balancing one.

A few practical things follow from that.

1. Build critical mass locally. How densely your early adopters are clustered in pockets matters far more than the overall percentage. A change that has reached 10% of the organisation but with all of those adopters clustered tightly in a few teams is far more likely to tip than one at 20% spread evenly. So concentrate your seeding rather than spreading it thinly; build critical mass locally, where the new behaviour can become visibly normal to everyone nearby, before you widen the net.
2. The most powerful change agents aren’t always the most senior. As Battilana and Casciaro’s NHS work showed, the people best placed to seed a reinforcing loop often sit some way down the formal hierarchy; what matters more is a network position that makes them visible and credible¹¹. More specifically, change agents whose contacts weren’t well connected to each other did better at driving changes that departed a long way from the status quo (though they may be less useful for smaller, less disruptive ones). So map the informal network and seed at the influential points – people with cross-functional responsibilities (matrix reporting lines, collaborative roles) or in administrative roles (PAs, team leaders, admin support, office managers) are a good place to start.
3. Build the new loop at the same time as weakening the old one. Don’t just argue against and old loop without offering an alternative. Both loops need attention at once: strengthen the new one through concentrated seeding while weakening the old one by stripping out the structural supports that keep it steady.

Checklist

• Systems Thinking explains what feedback loops are, and how they can be both levers and blockers for change
• Balancing loops are one of the main things keeping systems stable and blocking change
• Reinforcing loops are one way of flipping a system into a new state
• You need to tackle both at once – neither will get you there on its own

Footnotes

1.  Kotter, J. P. (2012) Leading Change. Boston, MA: Harvard Business School Press; Hiatt, J. M. & Creasey, T. J. (2012) Change Management: The People Side of Change. 2nd edn. Loveland, CO: Prosci Learning Center Publications. ↩︎
2. Senge, P. M. (2006) The Fifth Discipline: The Art and Practice of the Learning Organization. Rev. edn. London: Random House Business; Meadows, D. H. (2008) Thinking in Systems: A Primer. White River Junction, VT: Chelsea Green. Sterman, J. D. (2000) Business Dynamics: Systems Thinking and Modeling for a Complex World. Boston, MA: Irwin/McGraw-Hill. ↩︎
3. Meadows, D. H. (1999) ‘Leverage points: Places to intervene in a system’, The Sustainability Institute (working paper); Meadows, D. H. (2008) Thinking in Systems: A Primer. White River Junction, VT: Chelsea Green. ↩︎
4.  Salganik, M. J., Dodds, P. S. & Watts, D. J. (2006) ‘Experimental study of inequality and unpredictability in an artificial cultural market’, Science, 311(5762), pp. 854–856. ↩︎
5. Milgram, S. (1963) ‘Behavioral study of obedience’, Journal of Abnormal and Social Psychology, 67(4), pp. 371–378. ↩︎
6. Bikhchandani, S., Hirshleifer, D. & Welch, I. (1992) ‘A theory of fads, fashion, custom, and cultural change as informational cascades’, Journal of Political Economy, 100(5), pp. 992–1026; Bikhchandani, S., Hirshleifer, D., Tamuz, O. & Welch, I. (2024) ‘Information cascades and social learning’, Journal of Economic Literature, 62(3), pp. 1040–1093. ↩︎
7. Centola, D., Becker, J., Brackbill, D. & Baronchelli, A. (2018) ‘Experimental evidence for tipping points in social convention’, Science, 360(6393), pp. 1116–1119. ↩︎
8. Sterman, J. D. (2000) Business Dynamics: Systems Thinking and Modeling for a Complex World. Boston, MA: Irwin/McGraw-Hill. ↩︎
9. For the underlying network effect, see Braess, D., Nagurney, A. & Wallacher, T. (2005) ‘On a paradox of traffic planning’, Transportation Science, 39(4), pp. 446–450. ↩︎
10. Centola, D. & Macy, M. (2007) ‘Complex contagions and the weakness of long ties’, American Journal of Sociology, 113(3), pp. 702–734. ↩︎
11.  Battilana, J. & Casciaro, T. (2012) ‘Change agents, networks, and institutions: A contingency theory of organizational change’, Academy of Management Journal, 55(2), pp. 381–398. ↩︎